Improving Our Understanding of Seabird Behaviour at Sea

4 Discussion

This study aimed to investigate how weather and diurnal patterns influenced the behaviour and distribution of seabirds at sea during the breeding season. A particular focus for these analyses was the extent to which data collected during boat and/or digital aerial surveys may be biased as a consequence of the constraints imposed in relation to when such surveys can be carried out (e.g. during the day and sea-state 4 or lower). By using seabird tracking data, we were able to consider how the behaviour and distribution of five key species (Kittiwake, Lesser Black-backed Gull, Gannet, Guillemot and Razorbill) varied over the course of the day and in relation to wind speed and direction. By considering this in relation to the constraints imposed by boat and digital aerial surveys, we are able to better understand the implications of relying on survey data in the assessment of the impacts from offshore wind farms, in particular, the implications for the assessment of collision, displacement and barrier effects.

As an initial finding, the use of high wind speeds (based on cut offs of > 8m/s and > 10m/s) appeared to be generally lower than the wind speeds "available" to individual birds when at sea. This indicates that birds away from the colony were actively avoiding areas with high wind speeds. This was more pronounced for Lesser Black-backed Gull and to a lesser extent, Kittiwake. Lesser Black-backed Gull, however, has a relative bias towards offshore use during the chick-rearing period in June and July when wind speeds tend to be lower compared to earlier in the breeding season. There was a notable exception to this pattern, with gannets breeding on Alderney appearing to actively select areas with higher winds that are associated with lower flight costs in this species (Amélineau et al. 2014). These general findings, however, should still be considered indicative; a full habitat resource-selection analysis would be needed to verify these patterns statistically.

4.1 Model parameters

Across the five study species, behaviour was classified based on a combination of step length and turning angle. Our analyses identified three core behaviours for each species – resting on the sea surface, commuting and foraging. For Gannet, Lesser Black-backed Gull and Kittiwake the floating state was characterised by short step lengths and relatively acute turning angles implying birds were moving over relatively short distances in a consistent direction. Commuting behaviour was also characterised by birds moving in a consistent direction, but over longer distances. Foraging behaviour was characterised by wider turning angles and shorter step lengths than those observed during commuting. Patterns in behaviour such as this are consistent with area-restricted search whereby birds travel directly towards areas with predictable resources before using slower, less directed flight when looking for foraging opportunities within those areas, with such flights potentially being intermittent in nature punctuated by attempts to capture and handle prey (Hamer et al. 2009; Votier et al. 2013; Sommerfeld et al. 2016). For Lesser Black-backed Gull and Kittiwake we also identified a fourth behavioural classification which was characterised by short step lengths and wide turning angles and we believe is likely to reflect birds perching on objects at sea, for example buoys and other structures, confirmed for example with Lesser Black-backed Gulls in the Irish Sea.

In contrast to the gulls and Gannet, for Guillemot and Razorbill foraging behaviour was associated with shorter step lengths and more acute turning angles. This is likely to reflect birds foraging by diving from the sea surface rather than searching for prey from the air. For a subset of data we were able to validate predicted foraging areas using TDR data to confirm where diving behaviour took place.

4.2 Impact of covariates on behavioural state

All species exhibited a strong effect of the time of day on the likelihood of different states occurring, with floating on the sea being much more frequent at night and less common during the day (Table 40). Some species such as Lesser Black-backed Gulls have previously been found to differ diurnally in other key behavioural parameters important to collision risk, including flight heights outside of wind farms (Ross-Smith et al. 2016), and time within wind farms (Thaxter et al. 2018). However, to our knowledge, this is the first study to fully test how behaviours such as commuting, floating on the sea and foraging varies across a wider suite of species.

Wind speed had a marked effect on the probability of individuals of species transitioning between states or remaining within their existing state (Table 40). Wind speed was found to have variable influence on state transitioning in Razorbills and Guillemots, although commuting step lengths maintained a positive relationship with increasing wind speeds (assuming they had tailwinds) as birds were able to fly faster. Across species, a key finding was that Gannets and Lesser Black-backed gulls were more likely to transition from floating on the sea to foraging/searching for prey in greater wind speeds and with headwinds, with strongest effects in strongest headwinds – this suggests birds may be able to take advantage of strong wind conditions, for example for Gannets use of high wind speeds resulted in lower flight costs (Amélineau et al. 2014), and birds may adjust foraging time in relation to wind speed, maximising foraging time and reducing transition to resting. Spatially, the interactions with wind speed and direction may translate into a limitation of foraging range for some species particularly those that do not use dynamic soaring as a flight method, as has been suggested (e.g. Allerstam et al 2019). Further, the energetic costs of taking off from the water (e.g. following plunge diving for prey) are very high (Nudds and Bryant 2000) but, are reduced during stronger wind conditions when birds take off into the wind (Furness and Bryant 1996; Mullers et al. 2009).

4.3 Species behavioural characteristics

For each species, the proportions of time spent floating, commuting and foraging were broadly consistent across different colonies (Table 41). However, there were clear differences in the proportion of time spent in each activity between species. The analyses suggest that Gannets spend a greater proportion of their time floating on the sea surface than either Lesser Black-backed Gull or Kittiwake. However, the four state model for the two gulls split the foraging behaviour from the three state model into a foraging and perching behaviour, indicating that the three species may have spent similar proportions of their time resting. In contrast, the auks spent a far lower proportion of their time commuting. This is likely linked to a higher wing loading of these species associated with a higher cost of flight, in turn a product of adaptation of wing-propelled foraging strategies below water (Thaxter et al. 2010).

4.4 Species distributions

Our analyses highlighted clear spatial patterns in the areas used for resting, commuting and foraging across all species. Generally, the differences in the areas used for each activity were greatest between day and night, rather than in relation to different wind conditions (Table 38). However, the robustness of these conclusions may be limited by the lack of data from stronger wind conditions. Such limited data may mean that KDEs are patchier and, whilst similar areas are being used, the extent of any overlap may be reduced. Where there were differences in area use, these were generally most pronounced for resting behaviour. For example, there were suggestions from the analyses the Kittiwakes may be less likely to rest in areas further offshore during periods with high wind conditions.

4.5 Flight speed

At present, estimates of seabird flight speeds typically used in assessment for collision risk modelling have been derived using Ornithodolites or radar systems (Pennycuik 1987, 1997; Alerstam et al. 2007). However, it is unclear how representative these values are and there are concerns that some estimates are based on extremely limited data. For example, the estimate flight speed for Kittiwake from Alerstam et al. (2007) is based on just two tracks over a period of 660 seconds. Our analyses highlighted that for gulls and Gannet, wind speed and direction could influence bird flight speed (ground speed), thus matching previous findings in Gannet and Kittiwake (Lane et al. 2019, Collins et al. 2020). This emphasizes the importance of collecting measurements of flight speed from a range of conditions, rather than relying on data collected over a short time period which may not be representative of the conditions to which birds are exposed.

Our measures of flight speed were calculated by dividing the step length (the distance between two adjacent GPS fixes) and the intervening time elapsed between these points. Birds in flight may not travel straight between two such consecutive points. For example many seabirds use dynamic soaring, i.e. increasing wind speed over height, and slope soaring over waves, i.e. updrafts of wind over wave crests (Pennycuick 1982) to deploy an energy saving non-flapping flight and results in a characteristic small scale flight manoeuvres (Sachs et al. 2013). Hence any new flight speeds estimates presented here are likely to be a minimum value and further dependent on the time interval used (see Safi et al. 2013). The HMMs were used to differentiate what is classified here as commuting and foraging flight based on step lengths and the turning angles. The latter type of behaviour however, may also include other types of behaviour which lasted shorter than the sampling period covered. For diving species such as auks and to a lesser extent the Gannet, the particularly slow speeds recorded during apparent foraging flight (Table 42) will include time on the water between dives. The same is likely to be true for surface feeders such as Kittiwakes, and other gull species in open water, which are likely to spend time on the water between foraging bouts. Therefore, although we would urge caution in the interpretation and application of the flight speeds associated with foraging, this study highlights the need to distinguish between different types of flight behaviour and how this relates to flight speed. For Gannets the commuting flight speeds, estimated from Alderney and the Bass Rock, were broadly consistent with previous published estimates (Table 42 ; Pennycuik 1987). Whilst commuting flight speeds for Kittiwakes and Lesser Black-backed Gulls were slower than those used as part of current guidance (e.g. Alerstam et al. 2007), they were broadly consistent with the speeds estimated using laser rangefinders as part of the ORJIP Bird Collision Avoidance study (Skov et al. 2018). However, for both Razorbill and Guillemot, commuting flight speeds were considerably slower than those that have been estimated previously and, although these species are typically not considered vulnerable to collision (e.g. Furness et al. 2013), these estimates must nonetheless be treated with caution. However, for Gannet, Kittiwake and Lesser Black-backed Gull, commuting flight speeds were broadly consistent between colonies, suggesting that these values may be more widely applicable, for example, in relation to collision risk modelling.

¹Pennycuik 1987 ²Alerstam et al. 2007 ³Pennycuik 1997.

*see text for caveats regarding this data.

4.6 Implications for the assessment of collision risk

Flight speed is a key parameter in the Band Collision Risk Model (Band 2012). Firstly, it is used to estimate the total number of birds which may pass through a wind farm over any given time period, otherwise known as the flux rate. This is done calculating the length of time an individual bird may take to pass through the wind farm and scaling this up based on the density of birds within the wind farm at any given time. Secondly, it is used to estimate the probability of a bird passing through the rotor swept area of a turbine being hit by a turbine blade, based on the probability of the blade and the birds occupying the same space at the same time (Masden & Cook 2016). Consequently, estimates of collision risk are highly sensitive to assumptions about bird flight speed (Masden 2015).

As a consequence of the assumptions made by the Band Collision Risk Model (Band 2012), higher flight speeds results in a higher estimated collision rate. Hence using the previously published estimates of flight speed is likely to result in a higher collision rate than those derived as part of this study. This study also confirms the importance of accounting for different types of behaviour when assessing collision risk. Our analyses suggest that birds are likely to spend roughly equal proportions of time engaged in foraging and commuting flight which are likely to have differing flight speeds. Whilst the foraging flight speeds we estimated above should be treated with caution, we believe they highlight the importance of collecting estimates of flight speed that can be linked to behaviour. This can be achieved using existing GPS tags, which are capable of providing an instantaneous estimate of bird flight speed using the Doppler-shift principle for radio signals received from multiple satellites (Safi et al. 2013). Given the influence of wind speed and direction on bird flight speed accounting for these when estimating collision risk may have a significant effect on predicted collision rates. In particular, consideration should be given to incorporating different flight speeds in relation to upwind and downwind movements when estimating collision risk. For gulls and Gannet, there was a greater tendency for birds to be classified as commuting or foraging, as opposed to resting, during periods of stronger winds. This raises the possibility that by relying on survey data we may be underestimating the proportion of birds in flight during strong wind conditions. However, as data suggest that birds may avoid leaving their colonies during strong wind conditions, it is unclear what overall impact this may have on the total number of birds estimated to be at risk of collision.

There are also different levels of risk associated with different types of behaviour e.g. birds when foraging may be at greater risk if their attention is focussed on the sea surface, as opposed to when commuting flight, when they are more likely to be looking ahead (e.g. Martin 2011). Further work using cameras on birds (e.g. Votier at al. 2013 study on Gannet) would be extremely informative in testing this assertion. It is important to note however, that whilst we strive to understand how birds behave within wind farms and the likely risk of collision, the modelling framework does not currently allow us to easily include variation in flight speeds.

4.7 Implications for the assessment of displacement and barrier effects

Displacement and barrier effects are both likely to be manifested in a reduction in the density of birds within a wind farm. Whilst barrier effects are likely to result in an increased energetic cost as birds have to fly round a wind farm (Masden et al. 2010), displacement may reflect the loss of key foraging areas (Furness et al. 2013; Dierschke et al. 2016). The consequences of these two effects may be very different depending on the level of additional flight costs imposed by barrier effects and the relative importance of any lost foraging areas (Masden et al. 2010; Dierschke et al. 2016). Despite this, distinguishing between areas where birds may be vulnerable to barrier effects and those where they may be vulnerable to displacement using survey data can be extremely challenging, particularly in relation to species like Gannet which dive whilst in flight.

Our analysis highlights how tagging data can be used to link spatial information describing the distribution of birds at sea to their behaviour and differentiate between areas used for foraging, commuting and resting behaviours. When used in a pre-construction context, such information would enable us to assess both the likelihood and magnitude of any displacement and barrier effects more formally. For example, by comparing a wind farm footprint to areas used for commuting, it may be possible to infer the additional energetic cost associated with avoiding a wind farm. However, this would need to be considered in the context of prevailing wind conditions which may exacerbate or, ameliorate any additional costs. Similarly, in relation to displacement, it may be possible to assess the relative importance of any wind farm footprint to birds foraging from a given colony by adapting the approach in Wakefield et al. (2017) to estimate the total number of birds foraging within a particular area.

4.8 Transferability of results across colonies

Through our analyses we were able to identify the spatial and temporal distribution of seabird behaviours at different colonies. However, such data may not be available in relation to all colonies at which species may be impacted by offshore wind farms. Consequently, it is important to consider how transferable our conclusions may to other locations.

Our analyses suggested that the proportion of time that birds are engaged in different behaviours may be relatively consistent between colonies. However, there were clear differences in the locations and extents of areas used for different behaviours. Previous studies have shown that foraging areas may differ between colonies in relatively close proximity (Wakefield et al. 2013; Wischnewski et al. 2017). These effects may be influenced by intra-specific competition between birds from different colonies. However, colony-specific ecological and oceanographic conditions are also likely to play a key role in determining foraging areas and space use amongst seabirds. The geographic situation of a colony (e.g. mainland vs island, sheltered vs open coast) may play a key role linked to favourable hydrodynamic conditions and predictable resources such as upwelling zones, and shelf-sea fronts (Waggitt et al. 2016; Christensen-Dalsgaard et al. 2018; Grecian et al. 2012). Within colonies, there may be further partitioning in at sea space-use which may influence interactions with wind farms. While within-colony intra-specific variation is not investigated here, strong interspecific variation in movement direction can be seen at Puffin Island between Razorbills, flying northwest, and Guillemots, flying northeast (Appendix AD).

4.9 Analytical limitations

Although robust in the testing of the main aims of this project, the study has a number of limitations that need to be considered when drawing conclusions. First, as a result of the use-availability bias for some species in this study, there were limited sample sizes when making comparisons between low and high wind scenarios. Utilisation distributions are notably influenced by sample size in area (Soanes et al. 2013, Thaxter et al. 2017), which, therefore, hampers comparisons among KDEs. As such, for these reasons, the sizes of areas used under different scenarios are not presented and overlap indices should also be treated with some caution. Second, following from this, we also used an initial movement model to regularise the GPS data to precise equivalent specific time steps and HMMs can be used to explore the error in such interpolations and assumptions of movement between fixes through multiple permutations of the modelled tracks. However, given the volume of data across species and colonies analysed, such exploration of model error could not be carried out. Third, the amount of data itself was a limitation in this study – analyses were carried out at the colony level across years, being identified as a sensible compromise between model run-time and interest in the level of variation for the hypotheses tested. Note, however, for Bass Rock, the amount of data was large and as single model could not be reliably produced, instead individual year models were produced, which in turn provided opportunity to examine the variation across years for this case. Similarly, the amount of data collected for Lesser Black-backed Gulls spanned multiple months and years for individual birds, however, data for Lesser Black-backed Gulls were subset for investigation of offshore movements separate from terrestrial movements to meet the aims of the project. Fourth, within models, further aspects could be considered that were not investigated for the sake of simplicity – interactions for instance between time of day and wind speed were not tested, nor were the effects over distance to colony fitted in interactions to test if birds remained or transitioned between states further from the colony. Lastly, the study was focused on tracking data of known breeding individuals, yet at-sea surveys pick up other demographic components, such as non-breeders and immatures. The tracking datasets likely also contained failed breeders, particularly for Lesser Black-backed Gulls where nest monitoring prevents certain assessment of breeding status later in the season. Further, the stage of breeding was not tested in this study but could reveal further patterns. As such, the representativeness of the tracking dataset alongside birds that may be recorded in offshore surveys is here highlighted.

4.10 Recommendations for future work

Data collection

• The use of TDR devices in combination with positional telemetry technology such as GPS for diving species are recommended where possible. This would indicate key areas used for foraging.
• Further use of accelerometry incorporated with positional information collected by tags, or direct visual observations in wind farms would help further in refining behavioural states recorded at sea (Bouten et al. 2013).
• Additional weather covariates could also be explored to test correlations or biases in movement under different weather conditions, such as rainfall and visibility.
• Relating contemporaneous oceanographic variables e.g. state of tides or oceanographic fronts to positional telemetry data collected from tags
• Improved monitoring data at colonies which would allow differentiation between the different breeding stages of the bird as well as picking out birds that have failed that breeding attempt for the year.
• Determining the sex of the individual birds may also provide insight into their behaviour e.g. Sex mediated segregation of foraging area is well established in even monomorphic species e.g. Lewis et al. 2002.
• Deployment of tags pre-and post -construction of wind farms to provide greater understanding over how behaviour modified by the presence of artificial structures in the sea. This could arise through the provision of roosting or foraging sites for birds. There may also be changes in flight behaviour (e.g. height and speed) as birds actively avoid the turbine blades or even respond to changes in wind energy as consequence of the turbines rotating.

Analyses and modelling

• Further work examining the spatial distributions, for example in continuous time movement models (e.g. Wilson et al. 2018) could be beneficial for investigating spatial patterns in more detail.
• Incorporation of environmental covariates (see above under Data Collection) into predictive modelling frameworks. This would provide key insight into understanding the spatial and temporal patterns in birds' distribution and abundance at sea as well as their behaviour.
• It has already been shown that the flight height of gannets when foraging can put birds at a higher risk of collision with turbines compared to when commuting between sites (Cleasby et al 2015). Although flight heights were not part of this study, we would recommend that spatial patterns in flight heights for other species be further explored in future work.
• Revisions and/or updates to Collision Risk Models to better reflect spatial patterns in behaviour e.g. variation in key parameters such as flight height and speed.

4.11 Conclusions

Our analyses suggest that, in general, the at-sea distribution of birds during the breeding season was not strongly affected by wind speed or time of day. Consequently, when assessing the spatial overlap with offshore wind farms, data collected using boat or digital aerial surveys during the breeding season are unlikely to be biased relative to these variables. However, it is important to note that these results are limited to the breeding season when there were relatively few periods during which strong winds were likely to constrain at sea surveys. More data are needed in order to understand how strong winds outside the breeding season, when birds are not constrained by the need to remain close to their breeding colonies, may influence at sea distributions and the implications for survey data collected during these times.

Whilst survey data collected during the breeding season are likely to be representative of species distributions for breeding birds at this time, there are clear spatial patterns in behaviour which are more challenging to capture using at sea surveys. This is important as species behaviour may influence their vulnerability to the different impacts associated with offshore wind farms. For example, commuting birds may be vulnerable to barrier effects whilst foraging birds may be vulnerable to displacement. In order to better understand how offshore wind farms are likely to affect seabird populations and reduce uncertainty in the consenting process, a better understanding of seabird behaviour is required.